Interview: On the Importance of Classifiers

By Carl Smith on April, 9 2021


Stay up to date

You may have noticed we started a new blog, Variant Views, bringing you interesting news from around the world of human genetics research. We'll also be bringing you interviews from the VarSome community to help give a broader perspective on the issues and trends impacting our field. 

As VarSome's Application Scientist, Alex Joyner is an expert user and interpreter of our ubiquitous human genetic variant search engine and analysis platform.

He’s a big believer in the importance of recognized standards and is excited about the effect our upcoming CNV classifier could have.


VarSome: Hi Alex. Tell us a bit about yourself.

Alex Joyner: Sure. I’m the Application Scientist here at VarSome. That means I need to know our tools inside out so I can support our VarSome and VarSome Clinical users. Ultimately it’s about helping labs and clinicians get reliable information they can trust to support the clinical decisions they have to make. So I need to know about the 80+ databases we pull information from, how to understand that information and put it into context, and making sure our users are getting the most out of our ACMG and AMP variant classifiers.

In doing all of that, I spend a lot of time talking to our users - be they researchers or clinicians. Not only does it mean I can help them a little better, but it also means I can link back in with our development team. User insight is something we’re big on here at VarSome. The platform is under constant improvement, so it’s good for us to make sure we’re always being led by the users and finding ways to help them solve more challenging cases, faster.


V: You mentioned ACMG and AMP classifiers. Can you tell us a little bit more about those?

AJ: I’d love to! I think most people know VarSome as ‘THE variant search engine’, but I think that might change soon as we become known as ‘the classification experts’, and ultimately ‘the interpretation experts’. All of us in and around the field of genomics have seen a tremendous explosion of data over the past 5-10 years. The sequencing technology has gotten so fast and is getting more affordable… It’s truly turned this into a ‘big data’ field, and that is attracting a lot of amazingly innovative companies to take a look. When you have Google, Amazon, and all of these tech giants getting involved, you know it’s big. But, the problem is what to do with the data. There are all sorts of cool things you can do, but it’s for novelty’s sake unless we can improve human health in the clinic. And that’s where classifiers are so important.

ACMG’s variant interpretation standards were a huge deal back in 2015, and then AMP put out their guidelines for cancer variant classification. It might seem small, but what this means is that we can take all of that data and start applying it uniformly around the world in a clinical setting. Not only does it help clinicians make sense of the data, but it’s also an important part in building confidence in genomic data as clinically accepted evidence.

Geneticists have been annotating variants for years and years. Now with the classifiers, we’re moving into a much more active process by taking all of those annotations and turning them into something really useful rather than just labels.


V: How do you mean?

AJ: This is really at the heart of what VarSome is and does. If you think of an annotation, it really is just that; it’s a label and it’s in isolation. I mentioned earlier that we pull in data from over 80 databases right? Those are a lot of individual labels. So what we do through VarSome, is pull all of that information into a single place and apply the classifiers to tell you what it means in relation to the variant or variants you’re interested in. Imagine you’re in a forest - it’s the difference between looking at each tree individually and feeling overwhelmed, and then being able to have someone lift you up and show you the whole forest.


V: I see. But what happens when there are discrepancies with the classification of a variant?

AJ: Sure. It can happen. But not very often. What you have to remember is that we’re talking about a lot of different genes and pathways - thousands and thousands of them. Classifiers apply a set of standards and guidelines to help you get to a consensus answer based on criteria that have to apply to all cases as best they can. And they apply exceptionally well. That being said, it is possible that someone may disagree with a classification. And that’s one of the best things about VarSome.

As much as it’s a platform, it’s also a community project. In fact, our developers are working on implementing some great new features to support community interaction. But if you look at who’s using VarSome, it’s over 300,000 users around the world. Within that you have researchers who have devoted their careers to studying specific genes and pathways. So if they disagree with a particular classification, they can usually point you to a functional study that shows you the detail at a molecular level of what’s going on. And that is a very useful connection to be able to make if you’re trying to work through a particularly tough case.

It is worth pointing out though, that the discrepancies we’re talking about here are usually the difference between a variant being classified as ‘Pathogenic’ versus ‘Likely Pathogenic’, rather than ‘Pathogenic’ versus ‘Benign’!


V: Ok, so now we’re turning data into information. There’s a feeling out there that whole genomes have too much data that isn’t that useful in a clinical setting. What’s your take on that?

AJ: That’s a complicated question and one that’s very dependent on context. Sure, there are some things you don’t need to look at the whole genome to figure out. If you’re looking at a single gene or group of genes that are well defined for a particular indication, then you probably don’t need to go beyond a gene panel. But that’s in a clinical setting. In a research setting, you never know what you might find when you get into GWAS or PheWAS with all the other types of data that can be linked together these days.

That being said, let’s pull this back to impact. And for me that’s always going to come down to how we can help clinicians do more with what’s available to them today. The ACMG and AMP classifiers are extremely useful but you’re still just looking at exons and relatively simple point mutations. The human genome is a huge molecule. That means it’s going to be subject to a lot of structural variation. Thanks to all the amazing research that’s come out over the past decade or so, we know that structural variation can play a huge role in gene expression and disease. Even quite a rare event like a fusion is very relevant to cancer.

The biggest opportunity to really start putting all of that whole genome data to good use is copy number variants, though. With ACMG publishing a set of guidelines last year we’re about to implement those as a CNV classifier in VarSome later this month.

Having standards for this kind of thing is huge because it brings that element of consistency to analysis that is so important to clinicians. Without those guidelines, no one really knows what to do with all of that information in a clinical context. There is a lot of CNV data sitting out there uninterpreted - the CNV classifier could turn that into very useful information very quickly! So getting back to whether or not whole genomes are too much - I think we’re going to be getting more useful analyses from existing WGS data. I’m very excited about that!

We all know about the seemingly never ending diagnostic odysseys some families go through - anything we can do to reduce that is huge, as it’s the lack of an answer that can be so tough to deal with.


V: That’s very exciting news! Is there anything else you’d like to mention before we let you go?

AJ: It really is, right? This is stuff I love talking about, which is why I love being an application scientist. You get drawn into these great conversations with people trying to solve real problems. I’m going to be over, well virtually at least, at ACMG later this month and we’re exhibiting at AACR this year as well - if you want to talk interpretation, I want to hear what you have to say!

Register for Alex's Classification Tutorial

If you would like more information on any of what we do, we have some great resources on our website including a couple of whitepapers I’ve put together. And as always, get in touch with us - we always love to hear from VarSome users!

Submit a Comment

Stay up to date